Prediction of total bed material load for rivers in Malaysia: A case...

Environ Fluid Mech (2011) 11:307–318DOI 10.1007/s10652-010-9177-9

ORIGINAL ARTICLE

Prediction of total bed material load for riversin Malaysia: A case study of Langat, Muda and KurauRivers

Aminuddin Ab. Ghani · H. Md. Azamathulla ·Chun Kiat Chang · Nor Azazi Zakaria ·Zorkeflee Abu Hasan

Received: 19 February 2010 / Accepted: 2 June 2010 / Published online: 16 June 2010© Springer Science+Business Media B.V. 2010

Abstract A soft computational technique is applied to predict sediment loads in threeMalaysian rivers. The feed forward-back propagated (schemes) artificial neural network(ANNs) architecture is employed without any restriction to an extensive database compiledfrom measurements in Langat, Muda, Kurau different rivers. The ANN method demonstrateda superior performance compared to other traditional sediment-load methods. The coefficientof determination, 0.958 and the mean square error 0.0698 of the ANN method are higherthan those of the traditional method. The performance of the ANN method demonstratesits predictive capability and the possibility of generalization of the modeling to nonlinearproblems for river engineering applications.

Keywords Alluvial channels · Artificial neural network · Total-sediment load ·River engineering · Sediment transport

1 Introduction

Sand and gravel have long been used as aggregate for construction of roads and building.Today, the demand for these materials continues to rise. In Malaysia, the main source of sand

A. Ab. Ghani · H. Md. Azamathulla (B) · C. K. Chang · N. A. Zakaria · Z. A. HasanRiver Engineering and Urban Drainage Research Centre (REDAC), Universiti Sains Malaysia,Engineering Campus, Seri Ampangan, 14300 Nibong Tebal, Penang, Malaysiae-mail: [email protected]; [email protected]

A. Ab. Ghanie-mail: [email protected]

C. K. Change-mail: [email protected]

N. A. Zakariae-mail: [email protected]

Z. A. Hasane-mail: [email protected]

123

308 Environ Fluid Mech (2011) 11:307–318

is from in-stream mining. In-stream sand mining is a common practice because the mininglocations are usually near the “markets” or along the transportation route, hence reducingtransportation costs.

In-stream sand mining can damage private and public properties as well as aquatic habi-tats. Excessive removal of sand may significantly distort the natural equilibrium of a streamchannel. By removing sediment from the active channel bed, in-stream mines interrupt thecontinuity of sediment transport through the river system, disrupting the sediment mass bal-ance in the river downstream and inducing channel adjustments (usually incision) extendingconsiderable distances (commonly 1 km or more) beyond the extraction site itself.

In recent years, rapid development in Malaysia has led to an increased demand for riversand as a source of construction material, which has resulted in a mushrooming of riversand mining activities that have given rise to various problems that require urgent actionby the authorities. These include riverbank erosion, riverbed degradation, river buffer zoneencroachment and deterioration of river water quality. Very often, over-mining occurs, whichjeopardizes the health of the river and the environment in general. This study summarizesthe results based on field data collected at three river catchments in Malaysia, i.e., the RiverLangat, the River Muda, and the River Kurau. Fieldwork on selected sites for the three riv-ers was performed to assess the capacity of the river to convey both water and sediment.Data collection on the bed material was used to characterize the physical characteristicsof the sediment responsible for sediment transport, which determines the river response interms of erosion and deposition. The three rivers clearly have bed material sizes in the sand-gravel range based on the collected data in the present study [13]. This study shows that themeasured load can be predicted accurately for Malaysian rivers using the neural networksapproach.

The neural networks approach has been applied to many branches of engineering sciences.This approach is becoming a valuable tool for providing civil and hydraulic (river) engineerswith sufficient details for design purposes and river-management practices. Motivated bysuccessful applications in modeling nonlinear system behavior in a wide range of areas.

ANNs have been applied in hydrology and hydraulics [14,19]. ANNs have been used forrainfall-runoff modeling, flow predictions, flow/pollution simulation, parameter identifica-tion, and modelling nonlinear/ input–output time series [4]. Jain [18] used the ANN approachto establish an integrated stage–discharge–sediment concentration relation for two sites onthe Mississippi River in the United States.

Through Jain’s study, it was shown that the ANN results were better than those obtained bythe conventional technique [18]. Cigizoglu [9,10] employed ANNs to estimate suspended-sediment concentrations and made a comparison between ANNs and sediment rating curvesfor two rivers with very similar catchment areas and characteristics in northern England. Heused only discharge and sediment concentration parameters in his study. He showed that theestimates obtained by the ANNs were significantly superior to the corresponding classicalsediment rating curve. Yang et al. [24] used ANN to evaluate total sediment load formu-lae (where—river and location). Nagy [21] estimated that the natural sediment discharge inrivers in terms of sediment concentration by ANN model yields better results compared toseveral sediment-transport formulas [3,15,20,23]. Cigizoglu and Kisi [11] developed mod-els of ANN to estimate suspended-sediment. Tayfur and Guldal [22] estimated daily total-suspended sediment in natural rivers by ANN and a nonlinear black box model based upontwo-dimensional unit sediment-graph theory (2D-USGT) from precipitation data. The com-parison of results revealed that the ANN has a significantly better performance than the2D-USGT. Recently, Azamathulla et al. [7] used an ANFIS-based approach for predictingthe bed load for moderately sized rivers in Malaysia.

123

Environ Fluid Mech (2011) 11:307–318 309

2 Study area

This study covers three rivers, i.e., the River Langat, the River Muda, and the River Kurau,that have different levels of sand mining activities. River Langat recently has been a majorsource of sand for construction with the development of Putrajaya. River Muda has a longhistory of sand mining activity along the upper reach. Less sand mining is ongoing in theRiver Kurau upstream of the Bukit Merah reservoir.

2.1 River Langat

The River Langat basin occupies the south and southeastern parts of the State of Selangorand small portions of Negeri Sembilan and Wilayah Persekutuan. The basin is bounded onthe east by the Main Range and the Straits of Malacca on the west. The basin has diversetopography ranging from mountainous areas in the northeast, low rolling hilly areas in themiddle to lowlands in the south-west part of the basin. The geographic location of the basinis shown in Fig. 1(a).

The river system flows through the State of Selangor, Negeri Sembilan, and the FederalTerritory of Putrajaya. The main river, the River Langat, has a total length of about 180 km,and it forms one of the four major river systems in the State of Selangor. The River Langatbasin (the basin) has a total catchment area of 2,350 km2 and an average annual flow of35 m3/s, and the mean-annual flood is 300 m3/s. Use of River Langat is not limited to watersupply and includes other purposes such as recreation, fishing, effluent discharge, irrigationand even sand mining.

2.2 River Muda

The River Muda is the longest river in the state of Kedah, and it is situated in northernPeninsular Malaysia with its origin in the northern mountainous area of the state adjoiningThailand, as shown in Fig. 1(b). The basin has a drainage area of approximately 4,210 km2.However, small portions of the catchment lie within the upper boundary of the State of PulauPinang. In terms of administrative boundaries, the upper and middle reaches of the basinbelong to the State of Kedah, while the downstream of the river forms the boundary betweenthe states of Kedah and Pulau Pinang.

The main channel of River Muda has a length of about 180 km with a slope of 1/2,300(or 0.00043 m/m) from the river mouth to Muda Dam. The channel width is typically around100 m and widens up to about 300 m near the river mouth. The channel tends to erode dueto sand mining operations, aggravating bank erosion, and river total degradation. Per riversurveys in 2000, the shallowest point in the river is located 2.5 km upstream of the rivermouth, and it causes difficulty in navigation during low tides. At the upstream end of theRiver Muda is the Muda Dam, which acts as an extra storage for the Pedu dam. The twodams are part of the irrigation scheme. In general, the River Muda is still in its natural state,with riparian vegetation and flood plains made up of paddy fields along the riverbanks.

2.3 River Kurau

The River Kurau (Fig. 1c) represents the main drainage artery of the basin, draining an areaof approximately 682 km2, which is generally low lying. The river originates partly in theBintang Range and partly in the Main Range where the terrain in the upper reaches are steepand mountainous. The mid-valley of the river is characterized by low to undulating terrain,

123


Peninsular Malaysia

(b) River Muda

(a) River Langat

(c) River Kurau

Fig. 1 Study area

which gives way to broad and flat floodplains. Ground elevation at the river headwaters ismoderately high, at 1,200 and 900 m in Batu Besar and Batu Ulu Trap, respectively. Theslopes in the upper 6.5 km of the river average 12.5%, while those lower down in the valleysare much lower, on the order of 0.25–5%. The River Kurau basin is important as the mainwater resource for the Kerian Irrigation Scheme as well as the main domestic water supplyfor the Kerian District and the Larut and Matang District, State of Perak.

A dam was constructed 65 km upstream at the mid section of the rivers to form the BukitMerah Reservoir. This dam is operated principally to irrigate the paddy areas immediatelybelow the Reservoir. Upstream of the reservoir are two subsystems, namely the Kurau subsys-tem and the Merah River subsystem. Both drain through undulating to steep terrain. Areas inthe former subsystem were developed extensively for tree crop agriculture, while the PondokTanjong Forest Reserve forms the main land use of the latter subsystem. Largely rural in

123


Table 1 Range of field data for three rivers

Study area Study area

River Langat River Muda River Kurau

Q(m3/s) 2.75–120.76 2.59–343.71 0.63–28.94

V (m/s) 0.23–1.01 0.14–1.45 0.27–1.12

B (m) 16.4–37.6 9.0–90.0 6.30–26.00

Yo(m) 0.64–5.77 0.73–6.90 0.36–1.91

A(m2) 8.17–153.57 5.12–278.34 1.43–33.45

R (m) 0.45–3.68 0.55–3.90 0.177–1.349

So 0.00065–0.00185 0.00008–0.000235 0.00050–0.00210

Tb (kg/s) 0.027–0.363 0–0.191 0.080–0.488

Tt (kg/s) 0.2860–99.351 0.024–15.614 0.001–2.660

Tj (kg/s) 0.525–99.398 0.099–15.644 0.089–2.970

d50 ( mm) 0.31–3.00 0.29–2.10 0.41–1.90

Manning n 0.034–0.195 0.021–0.108 0.014–0.066

nature, the River Kurau basin has many riverine villages established from the mid to lowerreaches of the river.

3 River conveyance and sediment transport capacity

The data collection program for this study was implemented at three major rivers in Malaysiafrom 2007 until 2008 to assess the current state of river morphology based on on-site data andto determine the capacity of the river to act as it would naturally. Six study sites were chosenfrom each river for a detailed analysis of river conveyance and sediment transport capacity.

The surveyed cross sections for the River Muda and the River Langat are single threadchannels with the top width ranging between 22.5 and 134.0 m, representing medium-sizedrivers, and the top width for River Kurau ranges between 25.8 and 41.0 m, representing asmall-medium river. The slopes are between 0.00008 and 0.0021, indicating that the crosssections are still natural. The details of the morphological and hydrological descriptors andrange of field data are given in Table 1. The data collection includes flow discharge (Q), sus-pended load (Tt), bed load (Tb) and water surface slope (So). In addition, the bed elevation,water surface and thalweg measurement (the minimum bed elevation for a cross section) werealso determined at the selected cross sections. The total bed material load (Tj) is composed ofthe suspended load and bed load. The total bed material load must be specified for sedimenttransport, scour, and deposition analysis. Details of the measurement methodology are givenin [2]. The measured total bed material rating curves for these six sites at the three rivers areillustrated in Figs. 2–4.

4 Sediment transport equations assessments

A detailed sediment transport study at six sites for each river was conducted and it was foundthat Yang and Engelund-Hansen equations are able to predict the trend of sediment transportfor the three rivers.

123


0.1

1

10

100

0001001011

Tot

al B

ed M

ater

ial L

oad,

Tj

(Kg/

s)

Discharge, Q (m3/s)

Measured Data @ River langat

Equation 4

Present Study

Fig. 2 Comparison of River Langat sediment rating curve for this study and Eq. 4

0.01

0.1

1

10

100

1000100101

Tot

al B

ed M

ater

ial L

oad,

Tj

(Kg/

s)

Discharge, Q (m3/s)

Measured Data @ River Muda

Equation 4

Present Study

Fig. 3 Comparison of River Muda sediment rating curve for this study and Eq. 4

Yang [23] related the bed material load to the rate of energy dissipation of the flow as anagent for sediment transport. The theory of minimum rate of energy dissipation states thatwhen a dynamic system reaches its equilibrium condition, its rate of energy dissipation isat a minimum. The minimum value depends on the constraints applied to the system. Fora uniform flow of energy dissipation due to the sediment transport can be neglected. Yangequation for sand transport is:

123


0.01

0.1

1

10

1001010.1

Tot

al B

ed M

ater

ial L

oad,

Tj

(Kg/

s)

Discharge, Q (m3/s)

Measured Data @ River Kurau

Present Study

Equation 4

Fig. 4 Comparison of River Kurau sediment rating curve for this study and Eq. 4

log CT = 5.435 − 0.286 logWSd50

υ− 0.457 log

U∗WS

+{(

1.799 − 0.409 logWSd50

υ− 0.314 log

U∗WS

)× log

(V SO

WS− VCSO

WS

)}

(1)

where

Cv (ppm) = Ct (ppm)

SS

Critical velocity, VC is given by:

VC

WS= 2.5

log U∗V − 0.06

+ 0.06

Re∗ = U∗d50

V= 1.2 − 70

Vcr

WS= 2.05 for Re∗ ≥ 70

where Ct is total sand concentration (in ppm by weight), WS is terminal fall velocity, d50 isaverage particle diameter of granular material, υ is kinematic viscosity, U∗ is shear velocity,VS is unit stream power, and VCS is critical unit stream power required at incipient motion,Cv is sediment concentration by volume.

Engelund and Hansen [15] applied Bagnold’s stream power concept and the similarityprinciple to obtain the sediment transport equation below.

qs = 0.05ρsV 2[

d50

g (Ss − 1)

]1/2 [τ

(ρs − ρ) d50

]3/2

, (2)

where qs is total sediment discharge by weight per unit width, V is average flow velocity, S isenergy slope, ρ is density of water, ρs is density of sediment, d50 is median particle diameter,g is acceleration due to gravity, and τ is shear stress along the bed.

123


Table 2 Summary of discrepancy ratio for three rivers using Yang and Engelund-Hansen equations

Location Total of data Discrepancy ratio (DR) between 0.5 and 2.0

Yang equation Engelund-Hansen equation

Number of data % Number of data %

River Langat 60 30 50.00 31 51.67

River Muda 76 16 21.05 19 25.00

River Kurau 78 33 42.31 38 48.72

The assessment of two existing sediment-transport equations, the Yang [23] and Engelund-Hansen [15] equation, was performed after removing outliers for the 214 sets of data for thisstudy (Table 2). The assessment was based on the average size of the sediment (d50). Theperformances of the equations were measured using the discrepancy ratio (DR), which isthe ratio of the estimated load to the measured load (DR = estimated/measured). A discrep-ancy ratio of 0.5–2.0 (DR = 0.5–2.0) was used as a criterion in the evaluation of the selectedequations. The evaluation using these equations shows that all the existing equations, in mostcases, over-estimated the measured values, as shown in Table 2.

5 Multiple linear regression

Ab. Ghani [1] shows that good prediction of sediment transport in pipes could be obtainedfrom simple regression equations. It’s therefore decided to keep the form of the equation assimple and as easy to use as possible.

Based on dimensional analysis from previous works [1,21], the proposed function is givenas follows:

Cv = f

(V√

gd50 (Ss − 1),

R

d50,

B

yo

), (3)

where R is hydraulic radius, and B is water surface width. Utilizing all data from the threerivers in this study, the best equation is given as follows:

Cv = 2.42 × 10−5 ×(

V√2g (d50) (Ss − 1)

)0.022 (R

d50

)−0.2016 (B

yo

)0.104

. (4)

Figures 2–4 show the sediment rating curves for three rivers using Eq. 4.

6 ANN schemes

From the above analysis, we have attempted ANN—back propagation schemes for this study.Artificial Neural Network is architecture that fully connects elemental units, called neurons[16]. It draws its power, talent of learning, and generalization through the connectionistnetwork. The ANN is trained to provide a desired response to a specific stimulus (inputset). Generalization means that the trained and validated ANN model may produce log-ical outputs from independent inputs not used in the training and validation stages [16].The learning process might be supervised or unsupervised; the network may be feed for-ward or radial based, and so onwards. Hornik et al. [17] stated that multi-layer feed forward

123


Tj

weights

biases

.

.

.

B

V

Q

Yo

R

So

Fig. 5 Feed forward type NN

networks with as few as one hidden layer are indeed capable of universal approximationin a precise and satisfactory sense. They also concluded that if there is any lack of suc-cess in applications, the fact may arise from inadequate learning, insufficient numbers ofhidden units, or the lack of a deterministic relationship between input and output. In thisstudy, ANNs were developed for sediment data sets, the net work input –output as shown inFig. 5.

The present ANN structure consists of the input layer (with various numbers of inputs),one hidden layer, and the output layer. The feed-forward calculation starts at the input layer,moves forward to the next layer, and determines the output values for each neuron. Each unitor neuron in the network sums the weighted inputs and a bias passed from the previous layer,and then applies a nonlinear activation function to generate an output.

6.1 Selection of input parameters for the ANN

The selection of the input parameters is a very important aspect of neural network model-ing. In order to use ANN structures effectively, input variables in the phenomenon must beselected with great care. This highly depends on the better understanding of the problem. Ina firm ANN architecture, in order not to confuse training process key variables must be intro-duced and unnecessary variables must be avoided. For this purpose, a sensitivity analysis canbe used to find out the key parameters. Also sensitivity analysis can be useful to determinethe relative importance of the parameters when sufficient data are available. The sensitivityanalysis is used to determine the effect of changes and to determine relative importance oreffectiveness of a variable on the output. The input variables that do not have a significanteffect on the performance of an ANN can be excluded from the input variables, resulting ina more compact network. Then, it becomes necessary to work on methods like sensitivityanalysis to make ANNs work effectively [4]. The parameters that affect the total sedimentload can be given in a form as: Tj = f (Q, V, B, Yo, R, So) to establish an ANN architecture(Fig. 5).

6.2 Determination of ANN architecture

A total of 214 cross-section averaged load observations were divided into two parts. Seventypercent of the data was reserved for training, the rest for testing. The training data set was

123


Fig. 6 Observed versus predicted Sediment load ANN—BFG (inputs: eight hidden neurons)

used to determine the best weights and biases for the network. The testing data pattern wasused to measure the generalization performance of the selected model [16]. Data for each setwere selected randomly but statistical consistency between the sets was ensured. The divisionof data is a complex task [8]. The data from all rivers were uniformly distributed among thetraining and test data sets.

The logistic function, which is used in all layers of the proposed network, is constrainedbetween 0 and 1. Thus, the entire data set was rescaled so that one predictor did not dom-inate the model [12]. There are many different alternative standardization methods usedto create data with zero mean and unit standard deviation, but herein, for all ANN mod-els, the input(s) and the output were rescaled between 0.0 and 1.0. Researchers [5–7] usedANN techniques in modeling have used many different criteria to evaluate model perfor-mance. There is no generally accepted standard for the assessment of ANN performance. Acommon procedure is the use of the root-mean-square error (RMSE) and the coefficient ofdetermination R2 when evaluating the goodness of fit of models. After selecting the modeland the network type, the outputs predicted on the basis of the trained and the validatedweight bias matrices were compared with the three traditional total sediment load dischargemodels.

Because there is no specific method to determine the network structure, it was found thatthe best model representing the total load phenomena is in the form of the ANN (6,10,1). Thenumber of neurons in the hidden layers was determined through a trial-and-error approach.The ANN model was shown in Fig. 5 dimensional input parameters and one desired result,i.e., the sediment load. The model was trained using 150 data sets and tested with the remain-ing 64 data sets. The input 8, 1 hidden layer with 10 hidden nodes and 1 output, the trainednetwork yields satisfactory performance with R2 = 0.958 for BFG scheme (Fig. 6). Theother schemes were attempted and presented in Table 3.

123


Table 3 Error measures fordifferent ANN schemes

ANNschemes

Iteration number(Epoch)

Coefficient ofdetermination, R2

Mean squareerror

BFG 2000 0.958 0.070

CGF 1000 0.717 11.788

CGP 1200 0.694 12.567

OSS 600 0.746 10.457

LM 40 0.959 0.075

GDA 120 0.656 12.845

GDX 150 0.704 12.689

RP 200 0.813 5.678

SCG 80 0.876 1.234

CGP 1500 0.699 12.678

Yang [23] 0.722 10.376

Engelund–Hansen [15] 0.623 12.735

7 Conclusions

Total load transport in rivers is a complex phenomenon. The nature and motivation of tra-ditional total load models differ significantly. These approaches are normally able to makepredictions within about one order of magnitude of the actual measurements. To overcomethe complexity and uncertainty associated with total-load estimation, this research demon-strates that an ANN model can be applied for accurate prediction of total-load transport.A feed forward-back propagated (BFG) ANN model with one hidden layer with 10 hiddenneurons was found to perform adequately. The ANN model was able to successfully predicttotal load transport in a great variety of fluvial environments, including both sand and gravelrivers. Also, the ANN prediction of mean total load was in almost perfect agreement with themeasured mean total load. The high value of the coefficient of determination (R2 = 0.958)implies that the ANN model provides an excellent fit for the measured data. These results sug-gest that the proposed ANN model is a robust total load predictor. This study demonstrates asuccessful application of the ANN modeling concept to total load sediment transport. Despitehaving five index parameters, the conclusion that only eight parameters are required to pre-dict total load, is in agreement with previous works [24]. The value of the ANN approachis that the nonlinear function need not be the same for all fluvial environments. The geneticprogramming will be used to predict sediment load in the future with more database.

Acknowledgements The authors wish to thanks to Robert D. Jarrett, U.S. Geological Survey (USGS) forhis suggestions in preparation of this manuscript. The authors wish to express their sincere gratitude to Uni-versiti Sains Malaysia for funding a research university grant to conduct this on-going research (PRE .1001/PREDAC/811077).

References

1. Ab. Ghani A (1993) Sediment transport in sewers. PhD thesis, University of Newcastle upon Tyne, UK2. Ab. Ghani A, Zakaria NA, Abdullah R, Mohd Sidek L (2003) Guidelines for field data collection and

analysis of river sediment. Department of Irrigation and Drainage Malaysia, Kuala Lumpur, 35 pp. ISBN:983-3067-03-4

123


3. Ackers P, White WR (1973) Sediment transport: new approach and analysis. J Hydraul Div ASCE99(HY11):2041–2060

4. ASCE Task Committee (2000) Artificial neural networks in hydrology. I. Preliminary concepts. J HydraulEng ASCE 5(2):115–123

5. Azmathullah HMd, Deo MC, Deolalikar PB (2005) Neural networks for estimation of scour downstreamof ski-jump bucket. J Hydraul Eng 131(10):898–908

6. Azmathullah HMd, Deo MC, Deolalikar PB (2006) Estimation of scour below spillways using neuralnetworks. J Hydraul Res 44(1):61–69

7. Azamathulla HMd, Deo MC, Deolalikar PB (2008) Alternative neural networks to estimate the scourbelow spillways. Adv Eng Softw 39(8):689–698

8. Bowden GJ, Maier HR, Dandy GC (2002) Optimal division of data for neural network models in waterresources applications. Water Resour Res 38(2):1–11

9. Cigizoglu HK (2002) Suspended sediment estimation for rivers using artificial neural networks and sed-iment rating curves. Turkish J Eng Env TUBITAK 26:27–36

10. Cigizoglu HK (2004) Estimation and forecasting of daily suspended sediment data by multi-layer per-ceptrons. Adv Water Resour 27:185–195

11. Cigizoglu HK, Kisi O (2006) Methods to improve the neural network performance in suspended sedimentestimation. J Hydrol 317:221–238

12. Dawson CW, Wilby R (1998) An artificial neural network approach to rainfall-runoff modeling. HydrolSci 43(1):47–66

13. Department of Irrigation and Drainage Malaysia or DID (2009) Study on river sand mining capacity inMalaysia. DID, Kuala lumpur, Malaysia

14. Dogan E, Yuksel I, Kisi O (2007) Estimation of sediment concentration obtained by experimental studyusing artificial neural networks. Environ Fluid Mech 7:271–288

15. Engelund F, Hansen E (1967) A monograph on sediment transport in alluvial streams. Teknisk Forlag,Copenhagen, Denmark

16. Haykin S (1994) Neural networks: a comprehensive foundation. MacMillan, New York17. Hornik K, Stinchcombe M, White H (1989) Multilayer feed-forward networks are universal approxima-

tors. Neural Netw 2:359–36618. Jain SK (2001) Development of integrated sediment rating curves using ANNs. J Hydraul Eng ASCE

127(1):30–3719. Kisi O, Yuksel I, Dogan E (2008) Modelling daily suspended sediment of rivers in Turkey using several

data driven techniques. Hydrol Sci J 53(6):1270–128520. Laursen EM (1958) The total sediment load of streams. J Hydraul Div ASCE 84(HY1):1530-1–1530-621. Nagy HM, Watanabe K, Hirano M (2002) Prediction of sediment load concentration in rivers using arti-

ficial neural network model. J Hydraul Eng ASCE 128(6):588–59522. Tayfur G, Guldal V (2006) Artificial neural networks for estimating daily total suspended sediment in

natural streams. Nordic Hydrol 37(1):69–7923. Yang CT (1972) Unit stream power and sediment transport. J Hydraul Div ASCE 98(10):523–567, pro-

ceeding paper 929524. Yang CT, Reza M, Aalami MT (2009) Evaluation of total load sediment transport using AAN. Int J

Sediment Res 24(3):274–286

123

Prediction of total bed material load for rivers in Malaysia: A case...

Documents

Transcript of Prediction of total bed material load for rivers in Malaysia: A case...